Forecasting cross-section of stock returns with realised moments

The study tests whether realised moments of stock returns (mean, variance, skewness and kurtosis) computed from daily returns over the last month, quarter and year can predict the 1-month cross-sectional stock returns of 40 US-traded liquid stocks in the period 1986-2019. The performed univariate regression analysis confirmed a statistically significant positive effect between all the realised moments, computed over the last quarter and year, and the future 1-month cross-sectional stock returns, while the 1-month realised moments proved to be mostly insignificant. Multivariate analysis, performed with Elastic Net Regression, has confirmed that investment strategies utilising information from realised moments were able to significantly outperform a random investment in the out-sample period 2004-2019.